Physical activity has long been recognized as benefitting health. Physical activity levels can also be an indicator of health. For these reasons, a variety of individuals and institutions are interested in monitoring physical activity. Over the past years, activity trackers have become an increasingly popular way for individuals to monitor their physical activity. An activity tracker typically includes an accelerometer as a means of detecting the physical activity of the individual wearing it. The accelerometer generates a series of outputs that vary according to the magnitude of acceleration the accelerometer detects. Outputs may then be converted into readings that correspond to a particular level of acceleration. A series of readings over time may be viewed as a signal.
Activity trackers use accelerometer signals to quantify physical activity in a variety of units including calories burned, distance traveled, and steps taken. Devices that count steps are referred to, generally, as pedometers, though accelerometer-based step counters are sometimes distinguished from pedometers that count the oscillations of a weight. The initiation of contact between a foot and the ground during a step causes a peak in the acceleration signal from an accelerometer coupled to the individual taking the step. However, not all peaks in the acceleration signal are caused by steps. Peaks that are not caused by steps may be referred to as “noise.” Noise may be caused, for instance, by the vibrations of a car or the lack of damping in the accelerometer itself.
One method of filtering out noise is to set an acceleration threshold. Peaks that exceed the threshold are presumed to be caused by steps while readings below the threshold are presumed to be noise. However, the boundary between acceleration caused by step activity and that caused by noise may change substantially depending on factors such as how fast an individual is walking or running, the elasticity of the surface on which the step is taken, and the hardness of the individual's shoe soles.
The graph in FIG. 1 illustrates the shortcomings of a single, fixed threshold. Two samples of accelerometer gait data are superimposed on top of each other. Each sample illustrates two steps taken by an individual. The sample with higher peaks (110) comes from a healthy individual. The sample with the lower peaks (120) comes from a patient after surgery. The time scale on the x axis is in 24ths of a second and the units of acceleration on the v axis are 1/64 of g (g=free-fall acceleration in Earth's gravity). Threshold 1 (130) would record both steps from the healthy individual but would not capture any steps from the post-surgery patient. Threshold 2 (140) records both steps from the post-surgery patient, but counts four steps from the healthy individual when only two steps were actually taken. Threshold 3 (150), just slightly lower than Threshold 2, also records both steps from the post-surgery patient, but counts only one step from the healthy individual. Thus, a fixed threshold does not accurately count steps from both the patient and healthy individual.